Rollback serving_output and add TODO

ROCm · Apr 28, 2022 · a4abfa6 · a4abfa6
1 parent 177684a
commit a4abfa6
Showing 1 changed file with 3 additions and 38 deletions.
diff --git a/src/transformers/models/clip/modeling_tf_clip.py b/src/transformers/models/clip/modeling_tf_clip.py
@@ -1401,42 +1401,7 @@ def call(
         return outputs
 
     def serving_output(self, output: TFCLIPOutput) -> TFCLIPOutput:
-        text_hs = (
-            tf.convert_to_tensor(output.text_model_output.hidden_states) if self.config.output_hidden_states else None
-        )
-        text_attns = (
-            tf.convert_to_tensor(output.text_model_output.attentions) if self.config.output_attentions else None
-        )
-        text_model_output = TFBaseModelOutputWithPooling(
-            last_hidden_state=output.text_model_output.last_hidden_state,
-            pooler_output=output.text_model_output.pooler_output,
-            hidden_states=text_hs,
-            attentions=text_attns,
-        )
-
-        vision_hs = (
-            tf.convert_to_tensor(output.vision_model_output.hidden_states)
-            if self.config.output_hidden_states
-            else None
-        )
-        vision_attns = (
-            tf.convert_to_tensor(output.vision_model_output.attentions) if self.config.output_attentions else None
-        )
-        vision_model_output = TFBaseModelOutputWithPooling(
-            last_hidden_state=output.vision_model_output.last_hidden_state,
-            pooler_output=output.vision_model_output.pooler_output,
-            hidden_states=vision_hs,
-            attentions=vision_attns,
-        )
-
-        output = TFCLIPOutput(
-            loss=output.loss,
-            logits_per_image=output.logits_per_image,
-            logits_per_text=output.logits_per_text,
-            text_embeds=output.text_embeds,
-            image_embeds=output.image_embeds,
-            text_model_output=text_model_output,
-            vision_model_output=vision_model_output,
-        )
-
+        # TODO: As is this currently fails with saved_model=True, because
+        # TensorFlow cannot trace through nested dataclasses. Reference:
+        # https://github.com/huggingface/transformers/pull/16886
         return output